Particulate nitrate (PN) is a critical component of fine particulate matter (PM2.5), especially during wintertime over agricultural regions. In recent years, the contribution of wintertime PN to PM2.5 mass over the Midwestern United States (MWUS), an agriculturally intensive region, has stayed relatively high over time (up to 40%). PN formation is controlled by nitrogen oxides (NOX = NO + NO2), ammonia (NH3), and volatile organic compounds (VOCs). To best control wintertime PM2.5 burden, it is critical to determine PN sensitivity to precursor gases. Despite increases in NH3 concentrations and the flattening of previously decreasing PN trends since 2013 over the MWUS, the factors controlling local PN formation are not well constrained. Satellite tropospheric column NH3/NO2 ratios have been shown to be effective in diagnosing PN sensitivity over East Asia, Europe, and the Eastern United States. Here, we expand this approach to quantify spatially and temporally resolved multidecadal wintertime PN formation sensitivity in the MWUS. Using GEOS-Chem sensitivity simulations and satellite observations from the Ozone Monitoring Instrument (OMI) and the Infrared Atmospheric Sounding Interferometer (IASI), we diagnose wintertime PN sensitivity to NH3, NOx, and VOCs from 2007 to 2023. The NOx-sensitive regime dominates over rural areas, and the NH3-sensitive regime dominates over major cities. VOCs do not control PN formation. Recently, the NOx-sensitive regime is becoming more prominent over the MWUS. NO2 satellite column densities increase from 2007 to 2017 by 0.7 ± 1.1% yr-1 and accelerate from 2018 to 2023 (7.0 ± 1.5% yr-1). NH3 satellite column densities also slowly increase by 0.8 ± 0.6% yr-1 from 2007 to 2017 and accelerate to 2.1 ± 1.2% yr-1 from 2018 to 2023. Despite the recent increasing trends for both precursor gases, PN is still highly NOx-sensitive over the MWUS because of the high abundance of NH3. Interestingly, trends in surface NO2 concentrations oppose satellite observations, decreasing by -1.8 ± 2.0% yr-1 from 2018 to 2023, suggesting the influence of aloft NOX emissions in satellite retrievals. Our work indicates that targeting NOX emissions over rural areas can reduce wintertime PN and PM2.5, while controlling NH3 is more effective over cities.

The theory of passive seismic interferometry proposes the extraction of coherent subsurface response by cross-correlating ambient noise recordings and stacking them over time. This process assumes that ambient noise sources are sufficiently random and isotropic. However, in real-world conditions, especially in tectonically complex or populated regions, noise sources are often directional, intermittent, or concentrated in space and time. This implies that the subsurface response could be multimodal, complicating the interpretation of the interferometric subsurface response. To conduct surface-wave tomography by approximating inter-station impulse response with the interferometric response, it is essential to identify a mode related to stationary-phase ambient noise sources. This allows accurate surface-wave dispersion estimates. Clustering of ambient noise correlations, prior to stacking, is an attractive way to isolate the contribution of stationary-phase sources. Algorithms such as K-means clustering are not robust; they are sensitive to signal-to-noise ratio (SNR) and often fail to produce stable clusters when correlations are weak. We propose a more robust generative modeling approach to clustering and enhance the SNR of ambient-noise correlations. We used symmetric variational autoencoder (SymVAE) with discrete latent variables to allow identification of the modes of coherent subsurface response. Furthermore, our framework allows for a nonlinear stacking of correlations associated with the identified mode, thereby improving the SNR. We utilized this method to process the ambient noise data from the Himalayas and evaluated its ability to improve the imaging of the Main Himalayan Thrust. In some situations, our method resulted in a higher degree of acausal-causal symmetry, which confirms that the identified mode is due to stationary sources. The accompanying figure illustrates some clustering results and the corresponding improvement in dispersion curves for a station pair with an inter-station distance of approximately 300 km.

In the United States, disaster recovery operates within a broader framework of federal insurance and post-disaster aid programs that are central to rebuilding efforts. Despite substantial federal investment in disaster recovery, the underlying drivers of post-disaster recovery outcomes remain insufficiently understood with no universally accepted framework for measuring community resilience and often constrained by data limitations particularly in capturing socioeconomic aspects of recovery. Remote sensing has emerged as an alternative tool for assessing recovery, offering high-resolution, timely, and spatially explicit data to the traditional methods such as surveys and interviews. This study utilizes Visible Infrared Imaging Radiometer Suite (VIIRS) Nighttime Lights (NTL) data as a proxy for socioeconomic activity to assess the recovery of residential properties following Hurricane Harvey. Statistical analysis incorporates environmental, socioeconomic, and parcel-level characteristics to evaluate the effects of the National Flood Insurance Program (NFIP) and Individual and Household Program (IHP) on a short-term recovery period. Findings indicate that NFIP participation significantly contributed to residential recovery throughout the period of recovery following the hurricane. In contrast, IHP assistance showed measurable impact in the initial periods of recovery, likely due to the severity of damage and the immediate prioritization of basic needs. However, by the second and third quarters, federal assistance began to support recovery, reflected in rising NTL radiance levels, signaling the gradual return of human activity and economic restoration. The results also highlight the need for rigorous evaluation of such programs, with an emphasis on their integration into long-term mitigation and resilience planning.

Cosmic Rays Neutron Sensing (CRNS) emerged among proximal sensors as a reliable option for filling the scale gap between point probes and remote sensing.CRNS consists in detecting ambient neutrons close to the land surface, generated by the flow of high-energy particles from space. By placing a non-invasive detector above ground, the albedo of ambient neutron is collected. Thanks to the strong absorption of epithermal neutron by hydrogen in water molecules, the neutron count rate can be converted into a value of Soil Moisture (SM) averaged in the large volume of soil probed by the neutrons.A single CRNS probe can effectively monitor SM in an area up to a dozen hectares and a depth up to 50 cm, which defines an intermediate spatial scale in between point measurements and remote sensing, while also granting continuous measurement and a sub-daily temporal resolution. The measurement provided by a single probe can be representative of a whole agricultural or environmental monitoring site. However, it is the synergy between different technologies across scales that can boost the overall value of measurement systems. Local measurements like point-scale probes and soil sampling campaigns, together with a deep knowledge of the site, allow to interpret and validate the CRNS observed dynamics and in turn validate hydrological modelling. On the other hand, the horizontal spatial scale of the CRNS footprint is fairly comparable with remote sensing pixel, opening the possibility to interpolate data between satellite passes and, if a CRNS network is available, between multiple ground validation points. We here report observations from a set of sites in Northern Italy where CRNS probes manufactured by Finapp were installed. The sites span a variety of landscapes, soils and land uses (including agricultural, vineyard, forest, high-altitude grassland) and are monitored by a variety of approaches including capacitive probes, manual sampling campaigns, hydrological modelling (CLM5 and HYDRUS-1D model), and remote sensing products (ERA-5-Land, SMAP). The observation and comparison of different study cases show how the integration of CRNS with the other techniques can provide a more complete and reliable understanding of soil moisture dynamics at a local scale, making it possible to extend these understandings to a regional scale.

A document by Matthew Penn. Click on the document to view its contents.

Alfalfa is a highly valuable and widely cultivated perennial forage crop in the United States, playing a crucial role in the food supply chain as a primary feedstock for livestock. For alfalfa producers, both yield and quality are key concerns, and the timing of harvest is a critical management practice to maximize yield, quality, and profitability. To address these needs, this project delivers AlfAdvisor, a free, publicly available cyber-platform that integrates satellite remote sensing, machine learning, and economic modeling. AlfAdvisor provides: (i) timely, field-scale estimations of alfalfa yield and quality, and (ii) optimized harvest scheduling by accounting for essential factors such as the yield-quality trade-off, drying rate, and weather risks (e.g., rainfall on cut forage). Through an intuitive web interface, users can visualize yield and quality maps for their fields, view near-term yield and quality forecasts, and explore estimated yield and net revenue outcomes for different cutting dates. AlfAdvisor is poised to have an immediate impact on alfalfa producers, forage and livestock industry professionals, and researchers by enabling improved production and quality through data-driven management, thereby increasing economic opportunities for stakeholders. It is also expected to help improve alfalfa quality and availability in the livestock supply chain to increase animal and milk production, which in turn, will also benefit consumers by maintenance of low-cost animal products.

Accurate representation of land-atmosphere feedbacks is essential for improving the predictive skill of Earth system models on policy relevant timescales. A key component of these feedbacks is the coupling between soil moisture (SM) and evapotranspiration (ET), which controls surface water and energy exchanges. In this study, we examine whether SM–ET coupling in the Community Land Model (CLM6) can be improved through parameter tuning. To explore parametric effects on coupling, we developed a machine learning emulator that reproduces CLM output at a fraction of the computational cost. We trained a multi-layer perceptron on simulations from a CLM perturbed parameter ensemble to emulate annual root-zone SM and ET across space, time, and the parameter space of 56 plant traits and soil properties. The emulator accurately identifies regions where CLM underestimates SM–ET coupling relative to accepted benchmarks. Coupling strength responds differently to parameters across climate regimes: plant traits (e.g., maximum hydraulic conductance) control coupling in warm, dry regions, while soil traits (e.g., saturated volumetric water content) dominate in cold, wet areas. Arid regions show potential for improving coupling direction through parameter changes. In contrast, tropical forests exhibit persistent coupling biases that are insensitive to parameters, suggesting structural errors linked to limited plant functional types in biodiverse ecoregions and poor representation of complex canopies. Findings from this work can inform targeted model development and demonstrate how perturbed parameter ensembles and spatiotemporal emulators can be used to disentangle parametric versus structural sources of model error.

Land management and stewardship teams continue to lack the tools to capture 3D spatiotemporal insights of the ecosystems they oversee. For wildfire management at the wildland-urban interface, teams face challenges in capturing vegetation growth over time after a fuel reduction program and connecting seasonal changes to the vegetation distribution across the treated area. Current approaches rely on triangle meshes or point clouds generated from photogrammetry or LiDAR surveys on drones or hiked traverses. However, the difficulties in optimizing these meshes lead to large triangles that inadequately approximate the bulk vegetation shape, and the point cloud data is often too sparse for local plant-scale understanding. To address this gap, we extend recent machine learning-based computer graphics techniques in 3D Gaussian Splatting (3DGS) to reconstruct scenes at the wildland-urban interface from handheld imagery with sufficient detail to identify species, capture individual leaves, recover plant stature, and disambiguate overhanging plant individuals. We develop a new method to match the 3DGS reconstruction of these scenes across months, associating plant growth across seasons interactively in 3D. To achieve the centimeter-level matching, we adapt the Umeyama algorithm and the iterative closest point algorithm from point cloud maps to the 3DGS scene, leveraging the probabilistic interpretation of the 3D Gaussian data structure and robustly handling visual and geometric changes associated with vegetation phenology over time. We have applied our method to recent pile burns at Stanford’s Jasper Ridge ’Ootchamin ’Ooyakma Biological Preserve at monthly intervals. We demonstrate differences in ecological response where some piles featured the unexpected return of a rare and threatened bushmallow, and others remained more barren. This pile burn microcosm implicates the need for plant-level 3D spatiotemporal models to understand ecosystem recovery to fire mitigation practices.Please visit the project page for the spatiotemporal alignment video and more information: https://danineamati.github.io/burn-ecorecovery.github.io/

Small water reservoirs play a vital role in supplying local water demands, especially in arid and semi-arid regions. However, these reservoirs are highly vulnerable to evaporation losses [1]. Modular floating covers offer a practical and scalable solution to reduce evaporative water losses. While previous studies focused on cover properties and seasonal effects [2,3], the impact of broader climatic variations on evaporation suppression efficiency (ESE) remains understudied. This study uses a validated physically-based model [4] and MERRA-2 meteorological data to showcase the variation of ESE across diverse climatic regions in Iran. Preliminary results show that annual ESE changes between 75 to 81%. We observed a strong negative correlation between potential evaporation (PE) and ESE. While ESE in arid regions with high values of PE remains almost constant during the year, our analysis reveals sharper winter declines in ESE for humid regions with lower values of PE. This study provides a regional assessment of the water-saving potential of evaporation suppression floating elements and enables to identify areas where floating covers can be applied most effectively.

The unpredictable nature of the start and spread of fires prove them to be a prominent threat to cities across the United States. It is especially challenging to predict fires due to both the environment and humans acting as contributing factors to their emergence. Limited detailed knowledge on fire potential has led to poor resource allocation by cities, contributing to existing fires growing to uncontainable levels. This problem can be addressed using a holistic method to evaluate the likelihood of wildfires. Our goal was to create FIRECAST: a Fire Index Risk Estimator using Climate, Anthropogenic, and Soil Trends. This exploratory tool utilizes machine learning to provide timely risk assessments to evaluate a region’s susceptibility to wildfires. For inputs, we used data from the Global Human Settlement Layer, Normalized Difference Vegetation Index and Water Deficit from MODIS, as well as soil moisture, maximum temperature, precipitation, and wind speed. Furthermore, we utilized images of our local community collected through the GLOBE Observer App to visually validate land cover conditions and enhance the spatial relevance of our model inputs. We then preprocessed the data to ensure consistency and handle missing values. In order to train FIRECAST, we used XGBoost, a Gradient Boosting model known for its ability to handle nonlinear environmental data. Finally we used mean absolute error to evaluate its effectiveness. FIRECAST aims to provide a near on-demand wildfire risk calculation to aid local communities in resource allocation and preventative measures. It is clear that humans are a risk factor for wildfire potential and must be included in risk assessments. FIRECAST would provide authorities with more frequent risk estimations that take into account both environmental and anthropogenic factors. In the future, expanding input dataset variety to include more social factors, such as population density and even cultural fire-related practices, as well as historic wildfire data, would improve the accuracy of FIRECAST and ensure greater consistency with established indices including the FEMA National Risk Index and USGS Wildland Fire Potential Index. Improvements to FIRECAST will continue to further our objective of providing superior risk assessments to prevent and protect against wildfires.

Coastal surface currents influence processes like pollutant transport, sediment resuspension, and navigation safety, presenting the need for high-resolution monitoring tools for informed decision-making. We present a remote sensing approach based on unmanned aerial systems (UAS) that estimates surface currents using Doppler analysis of video collected from a continuous UAS flight transect over Freeport Harbor, Texas. This work extends the open-source MATLAB software CopterCurrents, originally developed for UAS videos collected by hovering at fixed station points (Streßer et al., 2017), to linear flights in which the continuous videos are spatially segmented into equivalent hovering videos. The segmentation is achieved by tracking and analyzing the motion of subwindows (240×240 pixels each) over the fixed ocean surface, each representing approximately 10×10 m2. Instead of relying on a fixed hover, each subwindow is tracked as it moves through the camera’s field of view during flight. For subwindows with sufficient temporal overlap (approximately 30 s duration, which is greater than 248 frames in this study), the mean 2D surface velocity field is extracted using three-dimensional fast Fourier transform and Doppler fitting to the linear, deepwater dispersion relation for surface waves (Streßer et al., 2017). Subwindow positions are geo-referenced in UTM coordinates based on UAS flight logs. To remove spurious current vectors, a signal-to-noise ratio threshold was applied to filter out noisy data, along with an upper velocity limit.We apply this method to a 10-minute UAS mission over Freeport Harbor, covering approximately 900 m and producing over 1000 geo-located subwindow measurements. The final output is a spatially continuous current map overlaid on satellite imagery, visualized through a vector field and a current-speed heatmap. This approach enables high-resolution current estimation without the need for hovering, thereby supporting efficient deployments for estuarine dynamics, rapid-response coastal monitoring, and operational nearshore oceanography.ReferencesStreßer, M., Carrasco, R., & Horstmann, J. (2017). Video-Based Estimation of Surface Currents Using a Low-Cost Quadcopter. IEEE Geoscience and Remote Sensing Letters, 14(11), 2027–2031. https://doi.org/10.1109/LGRS.2017.2749120

A critical gap in understanding the global methane (CH4) budget is the quantification of emissions from small artificial water bodies. One mechanism that may be driving increased CH4 emissions is an increase in natural emissions from inland aquatic ecosystems, such as small water bodies (<0.1 km²). The number, location, and seasonal extent dynamics of small water bodies—particularly artificial (or man-made) water bodies (i.e., irrigation or stormwater management ponds)—are still relatively unknown, despite their critical importance in CH4 emissions. Data from satellite imagery is a promising way to map the presence and areal extent of small water bodies, as they acquire data systematically over space and time. However, most prior research has used medium-resolution satellite imagery (30 m), which has been shown to underestimate the area of small water bodies. In this study, we employed machine learning (random forest) and deep learning (U-Net) to classify small water bodies in central and eastern North Carolina, using Sentinel-2 (10 m) and PlanetScope (3 m) optical satellite imagery. We then compared these classifications to a Landsat 30 m surface water product. We found that each resolution detected seasonal changes in surface water extent; however, the higher resolution imagery (i.e., PlanetScope) detected larger areas for the same water bodies. We also found that the higher-resolution imagery detected more small water bodies than the moderate-resolution imagery. Finally, we found that the higher resolution surface water extents had a stronger correlation with field-sampled CH4 concentrations from small artificial water bodies (R² 0.68) than the coarser optical imagery (R² 0.16). We conclude that higher resolution imagery can 1) identify more small water bodies than moderate resolution imagery; and 2) be used to provide more accurate estimations of local, regional, and global methane emissions. Such estimates can help countries better plan to meet their Global Methane Pledge.

In southeast Texas, abundant groundwater withdrawal has caused permanent aquifer compaction, resulting in land subsidence. With the region’s projected population growth and increased water demand, evaluating future subsidence impacts is crucial for effective water resource planning. This study presents a robust, scenario-based framework for evaluating future flood hazards and quantifying economic losses associated with inland subsidence in the Spring Creek Watershed, which spans across four counties in southeast Texas.

The 2017 GOES-R Post Launch Test (PLT) field campaign was conducted to validate the Advanced Baseline Imager (ABI) and Geostationary Lightning Mapper (GLM) instruments on board the GOES-16 satellite. As part of the campaign, a suite of airborne lightning and cloud measurements were collected from the high-altitude ER-2 aircraft, complemented by ground-based lightning and radar data. Of particular interest to an ongoing historical reanalysis of these data is (a) the characterization of optical measurements of lightning from the airborne Fly’s Eye GLM Simulator (FEGS) using ground-based detection of lightning radio emission from Lightning Mapping Arrays (LMAs) and (b) how these relationships vary with cloud and convective properties.The GOES-R PLT campaign included six ER-2 flights over an LMA, capturing lightning that occurred in multi-mode convection in varying regions across the continental United States. Multi-spectral optical measurements from the FEGS are characterized using the spatial and temporal characteristics of matched lightning flashes reconstructed from LMA detections. These relationships are then interpreted using cloud depth inferred from the airborne Cloud Physics Lidar as well as precipitation type and ice water content retrieved from ground-based polarimetric Doppler radars. These data are used to address how the interpretation of optical lightning data and its portrayal of lightning characteristics vary with cloud and convective context.

Over the past decades far-ultraviolet (FUV) remote sensing has played a key role in advancing heliophysics by providing valuable measurements of space plasmas and neutral species. However, current inversion techniques for retrieval of physical parameters such as neutral composition and temperature and plasma density from FUV emission data do not consider the Poisson nature of the data or incorporate spatial information. This leads to loss of information that can be leveraged to more accurately retrieve fields of physical parameters with spatial structure and quantify uncertainty. We propose here that by modeling the received photon counts as a permanental process and applying reproducing kernel Hilbert space techniques, we are better able to uncover structure in fields and reduce the effects of shot noise, allowing for more reliable estimates and accurate uncertainty quantification. We demonstrate the method by estimating thermospheric neutral temperature from FUV disk emission data from the NASA Global Observation of the Limb and Disk (GOLD) mission. Case studies with both simulated data during a geomagnetic storm in early November 2018 and actual GOLD L1C data products are presented.

Ground-level ozone (O3), nitrogen dioxide (NO2), and nitric oxide (NO) are key components of photochemical smog and well-established threats to public health and ecosystems. Because these gases interconvert within hours through photochemical reactions (e.g., NO + O3 → NO2 + O2), modeling them together lets information from each pollutant inform and improve the prediction of the others. However, regulatory monitoring networks remain sparse and uneven across space and time: the Contiguous United States (CONUS) hosts approximately 1000 O3 sites but only 400 NO2 and fewer NO stations, posing a challenge to the development of data-hungry deep learning models. We present a multi-task learning (MTL) framework that simultaneously estimates daily 1 km-resolution O3, NO2, and NO concentrations for CONUS. The architecture couples a shared spatiotemporal encoder—composed of a convolutional neural network (CNN) followed by a bi-directional long short-term memory (Bi-LSTM) network—with learned location and temporal embeddings; pollutant-specific decoders then produce individual pollution concentrations. Training leverages a comprehensive set of predictors: meteorology from Daymet (temperature, precipitation, vapor pressure, short‑wave radiation) and GridMET (wind speed and direction); chemical precursors from the National Emissions Inventory (VOC and NOx); land‑surface information from MODIS NDVI and a digital elevation model (DEM); reanalysis data from CAMS EAC4 and MERRA‑2; and satellite retrievals from Aura/OMI (total‑column O3, NO2, NO). MTL yields clear benefits over single-task baselines in all three pollutants, demonstrating that knowledge shared across chemically linked pollutants enhances generalization even for the well-monitored pollutant. The resulting 1km, daily O3-NO2-NO dataset from 2005 to 2021 will be made publicly available, providing an unprecedented resource for longitudinal epidemiological studies, exposure disparity assessments, and air quality management.

Reflectance spectroscopy has become a promising, non-destructive technique for capturing plant functional traits. Existing leaf spectral trait prediction models are trained on peak-growing season data, however, it is uncertain whether such models can fully capture trait temporal dynamics over the phenological cycle in seasonal environments. We monitored changes in traits and leaf spectra through the season and generated 7,515 fresh leaf spectra for 7 temperate species. We used partial least squares regression (PLSR) to develop three categories of trait models for leaf phenology based on trimmed spectra (400-2400 nm). We grouped the trait prediction models into complete leaf phenology and season-specific models, and assessed their performance on measured leaf traits. Our assessment of the predicted traits show that models trained with complete leaf phenology data have high accuracy for leaf mass per area (LMA) and equivalent water thickness (EWT) with R2 = 0.85 - 0.93 and %RMSE = 4.48 - 4.77, intermediate accuracy for nitrogen (N) with R2 = 0.64 and %RMSE = 10.69 - 10.71, and low accuracy for carbon (C) with R2 = 0.08 - 0.26 and %RMSE = 17.83 - 19.92. Season-specific models had low accuracy for LMA and EWT, with R² = 0.03-0.25 and %RMSE = 16.79-464.4, while C and N had poor accuracy, with R² = 0.00-0.14 and %RMSE = 22.59-89.88. Temporal variation in leaf phenology at intra- and interspecific levels was observed in traits and spectra. We show that not accounting for temporal variation in leaf phenology can bias our understanding of plant function. This study provides a path forward to mitigate potential biases in predicting functional traits using reflectance spectroscopy and machine learning.

The challenge of measuring the amount of water stored in mountain snowpack as SWE is usually addressed by a combination of computational modelling, meteorological observations and remote sensing. The best way to obtain ”true values” for validation is by in-situ coring campaigns performed by specialized personnel, which however suffers of critical limitations due to the accessibility of the sites. Recently, Cosmic Rays Probes (CRP) have emerged as a suitable candidate of proximal sensors deployable to remote areas, capable of providing continuous SWE measurements on site. The interaction of Cosmic Rays (high energy particles naturally flowing from space) with the atmosphere generates neutrons and muons that can be monitored as a non-invasive mean to probe the environment.Finapp developed a compact CRP characterized by the patented feature of being able to detect and discriminate both neutrons and muons with the same detector. This offers three different options to derive SWE from particle counts: - based on the drop of neutrons counted under the snowpack - based on the drop of muons counted under the snowpack - based on the drop of neutrons counted over the snowpack, thanks to the back-scattered neutrons These measurements are characterized by different horizontal footprints and saturation levels, with the former reaching up to 20 hectares by the third method and the latter up to 10000 mm SWE by the muons method. The extended footprint of CRP is of particular value to provide ground validation points for remote sensing products.We will present measurements obtained by these methods in the Italian Alps during the last two winter seasons, by a network of 25 Finapp systems for SWE monitoring (deployed by the Regional Environmental Protection Agency of Veneto - ARPAV) plus 2 more experimental installations (Colle del Nivolet and Cima Pradazzo). The measurements were validated by coring campaigns on site. We will compare the methods among them and with computational models, also highlighting the advantages of their simultaneous availability.